1

Toward the sense of touch in snake modular robotsfor search and rescue operations

Juan Gonzalez-Gomez1, Javier Gonzalez-Quijano1, Houxiang Zhang2, Mohamed Abderrahim1

1Robotics Lab. Universidad Carlos III de Madrid, Spain2TAMS, Department of Informatics, University of Hamburg, Germany

{juan.gonzalez.gomez, javier.gonzalez-quijano, Mohamed.Abderrahim}@uc3m.es,hzhang@informatik.uni-hamburg.de

Figure 1. Typical urban scenario after an earthquake. A highly unstructuredand chaotic environment with a lot of pieces of rumble. Likely, there arepeople inside that should be rescued immediately. City of Puerto Principe,Haiti. Photo courtesy of tumundovirutal.wordpress.com

Abstract—Snakes modular robots are good candidates forbeing used in Urban Search And Rescue (USAR) operationsbecause of their flexibility, good adaptation to the terrain andsmall section. We propose to design and build snake robots thatcombine three capabilities: locomotion, climbing and grasping.The last one allows the robot to remove objects for clearing thepath to the trapped people. The sense of touch is key to achievingthese three capabilities. To implement it, our novel approach isbased on the idea of touch rings and touch strips. In this papersome preliminary ideas are presented.

I. INTRODUCTION

Unfortunately disasters caused by earthquakes, floods,tsunamis or hurricanes are very often in the news. In thesescenarios the priority is to find survivals quickly, otherwisethey will die. In urban disasters the environment is extremelyunstructured, chaotic and full of pieces of rubble everywhere(fig. 1). This makes very difficult and dangerous for the USAR(Urban Search And Rescue) teams to move and locate people.Besides, it is clear that if a building has collapsed, there will belikely people inside that should be rescued as soon as possible.

During the last decade, researchers around the world havebeen facing the problem of designing tele-operated robots

capable of moving in such environments and to assist theUSAR teams. Properties such as versatility, adaptability andflexibility are taken into account when designing these robots.

Yim et al.[1] were the first to propose modular robotsfor this purpose, due to their three promises of versatility,robustness and low cost. Miller[2] suggested snakes robots(consisting of yaw modules with passive wheels) as goodcandidates for moving in unstructured terrains, in a similarmanner than real snakes do. Wolf et al.[3] from the Departmentof Mechanics at CMU developed a new concept consisting ofan elephant trunk-like robot mounted on a mobile base. Acamera located at the end of the trunk was used for inspectingunreachable areas such as small cracks and pipes.

The Japanese government is specially interested in USARapplications. They are funding and promoting research for thedevelopment of practical search-and-rescue robots. Prof. Hi-rose and their colleagues at the Tokyo Institute of Technologyhave been working on these robots for many years [4]. Theyhave developed robots with tracks and wheels, such as Souryu,Genbu, Kohga, Gunryu and the latest Helios VIII[5]. Theyhave also proposed the snake robots as candidates for USARapplications because of their advanced motion capabilities:their bodies can act as “legs”, “arms” or “fingers” dependingon the situation. Since mid 70’s they have been working onthe ACM snake-like robot family[6]. The ACM-R3 and R4have passive wheels larger than the profile of the link thatfully cover the whole body. This enables the snake to slidesmoothly among the rubble.

In the Institute of Robotics at Beihang University, Zhang etal. have developed the JL-I robot[7]. It is composed of threeidentical autonomous modules with 3 DOF joints. It can climbobstacles, stairs and recover itself in case of turning over.

One of the latest USAR robots is Amoeba-I[8] developedat the Shenyang Institute of Automation in China. It iscomposed of three tracked modules and has nine locomotionconfigurations. It can change its configuration automaticallyto adapt to the environment.

Modular robots with a 1D topology can be divided into twogroups: serpentine and snake robots. The former use activetracks or wheels for self-propulsion while the latter use bodymotions in a similar manner than their biological counterparts.A very successful serpentine robot is Omnithread developedby Granosik, Hansen and Borenstein at the Mobile RoboticsLaboratory in the University of Michigan. It is a novel design

2

Figure 2. A general scenario where there is a trapped person and the accessis blocked by pieces of rumble.

consisting of 5 segments with moving tracks on all their foursides to assure propulsion even when the vehicle rolls over.The latest prototype is OT-4[9]. Rimassa et al. [10] havedeveloped a serpentine robot with climbing abilities.

Even though serpentine robots are very promising for USARapplications, they lack the flexibility of real snakes and it is dif-ficult to attach to them artificial skins because of their wheelsor tracks. Snake robots, on the contrary, are similar to realsnakes and inherit some of their properties. Perambulator[11],developed by Ye et al. at the Shenyang Institute of Automationin China, is able to perform a powerful propulsion and a highmobility. Every segment contains a kind of omni-directionalwheel with free rollers that enables this robot to perform theserpentine locomotion (like real snakes).

Another modular robot with high locomotion capabilitiesis Hypercube[12], developed in our group. It can perform atleast 5 different gaits: moving in a straight line (forward andbackward), moving in a circular path, side-winding, rotatingand rolling.

Our goal is to develop a new snake modular robot forsearch and rescue applications. In this paper we propose apreliminary novel design of its tactile system that will be usedfor improving the robot locomotion as well as its climbing andgraping capabilities.

II. SNAKE MODULAR ROBOTS FOR SEARCH AND RESCUEOPERATIONS

Modular snake robots have a lot of potential for search andrescue operations. Consider the scenario shown in figure 1where there is a collapsed building (image of Puerto Principe,in Haiti). A USAR robot would be very helpful to exploreinside the house and searching for people. The are two possiblealternatives. On one hand, the robot can go through the barsof the window if its size is small enough. On the other hand,it can climb the wall and enter through the upper hole, but therobot should be flexible and have climbing capabilities. Snakerobots could meet these requirements.

Now let’s consider a general scenario as the one shown infigure 2 where there is a person trapped inside a house. First(fig. 2-1) the snake robot approaches the zone. Obviously,locomotion capabilities are needed. Second (fig. 2-2), thesnake arrives to a wall and has to climb it. Therefore, climbingcapabilities are necessary. Third (fig. 2-3), there are some

pieces of rubble blocking the access to the person. The snakeclears the path by grasping the objects and getting rid of them.Grasping capabilities are required. Finally (fig. 2-4), the robotreaches the person.

This scenario suggests that the robot should have at leastlocomotion, climbing and grasping capabilities. In previouswork, our group has researched on these three areas. Thelocomotion of snake modular robots of any length has beenwidely studied and implemented on real robots[13] usingsinusoidal oscillators that can be programmed with low cost8-bit micro-controllers. A novel inspired modular climbingcaterpillar that combines climbing techniques with locomotioncapabilities was designed and tested in [14]. For climbing,a low-frequency vibrating passive attachment principle wassuccessfully developed. Moreover, grasping capabilities arebeing studied and simulated[15].

The sense of touch is key to achieve these three capabilities.For locomotion, touch sensors provide the necessary feedbackfor enabling the robot to adapt its body to the terrain andtherefore moving more efficiently. In addition, situations inwhich the robot get caught on the ground can be detected. Forclimbing, the attaching forces can be measured to guaranteethat the robot will not fall down. Finally, the sense of touchin grasping allows the robot to apply the necessary force toremove objects from the path as well as optionally detectingthe shape of these objects.

III. SENSE OF TOUCH

A. IntroductionPerceiving the environment in animals needs multimodal

sensing capabilities. Humans combine sensory informationfrom different sources as touch, vision, and hearing. If notusing any of them, there is a gap in knowledge that makesdifferences between what is sensed and what is perceived[16].Simple experiments of exploring objects after putting handson an ice block for a while, therefore anesthetizing the skin,shows the difficulty of achieving stable grasp[17]. This occurseven when participants can see what they were doing.

There are several technologies and transduction principlesthat have been traditionally employed for the development oftactile sensing for robots, such as resistive, force, magnetic,optical, piezoelectric and capacitive based sensors[18]. Amongthem, capacitive sensors are commonly used in robotics appli-cations. Despite their two major drawbacks, stray capacity andsevere hysteresis, they are very small and sensitive. In additionthey can be easily arranged in dense arrays. As an exampleof their high degree of integration, Gray et al. developed in1996 an 8x8 capacitive tactile sensing array with 1 mm2 areaand spatial resolution at least 10 times better than the humanlimit of 1 mm. Schmidt et at. [19] designed a novel arrayof capacitive sensors for grasping purposes, which couple tothe object by means of little brushes of fiber. Maggliali etat.[20] proposed the novel idea of using a mesh of trianglesfor covering three-dimensional surfaces. They have designeda triangle module witch contains 16 capacitive sensors and allthe electronics.

Currently, it can be found commercially capacitive basedtouch array sensors such as the ones provided by ‘RoboTouch’

3

Figure 3. The touch rings approach for the implementation of the sense oftouch in snake robots

and ‘DigiTacts’ 1. Also, there exists various commercial ‘ca-pacitance to digital converter’ chips such as the AD7147 ’Cap-touch’ chip from Analog Devices. It have been successfullyemployed for the triangle module implementation[20].

B. Touch rings

Real snakes have a continuous skin, like the mammals, inwhich the sense of touch is distributed uniformly throughoutthe whole body. Designing an artificial skin similar to that ofthe snakes, with the same flexibility and equipped with touchsensors, is a big challenge. It still has not been realized, to thebest of our knowledge. In our approach, we propose the ideaof touch rings (figure 3) located at fixed positions along thebody axis. Therefore, the sense of touch is not continuous anduniformly distributed but discrete and concentrated in touchzones. Although it is not continuous, the distance between twoconsecutive rings (d) can be set according to the application.If higher touch resolution is required, this distance can bereduced. Thus, this approach is very versatile giving thedesigner the possibility of changing it.

In our opinion, USAR operations do not require a very highresolution. It should be high enough to allow the robot toperform the locomotion, climbing and grasping efficiently. Asa starting point we are planning to use one touch ring permodule.

C. Touch strips

Each touch ring consist of a flexible capacitive strip thatis bended (figure 4). One advantage of this design is that itcan be fitted into different snakes with different sections ofthe body trunk, due to the fact that the strip is flexible. In theexample of figure 3 the section is circular, but it can also beapplied to snakes with other section shape, like a square. Thestrip will fit regardless of the shape of the body section.

Therefore, our design is very versatile and it is valid for alot of different snakes. Moreover, the width and length of thestrip can also be changed according the application. The widerit is, the more capacitive sensors can be embedded.

1http://www.pressureprofile.com

Figure 4. The capacitive touch strips used for implementing the touch rings

D. Capacitive sensor principle

The touch strips are based on capacitive sensors. The stripis composed of three layers (figure 4). The total thickness isexpected to be less than 1mm. The first layer is a flexibleprinted circuit board (PCB) containing the electrodes for thecapacitors. The second layer is the dielectric. This materialshould be compressible and expandable. We are testing differ-ent candidates, such as silicone (S60) and polyester (PEPT)based ones. The third layer is the ground plane.

When normal forces are applied to the strip surface, thedielectric layer is compressed and the distance between theelectrodes and the ground plates is reduced. Therefore thereis a change in the capacitance of the capacitors that can bemeasured. The material in the second layer must recover itsinitial thickness when no forces are applied.

E. Electronics

Various commercial capacitive-to-digital converters chipsare available. In this work, we are using the MPR121 fromFreescale2 because of its low cost and I2C bus compatibility.Its main features are summed up in table I. It can measurecapacitance ranging from 10 pF to 2000 pF. Minimum size ofthe electrodes should be less than 0.25 cm, if high dielectricmaterials are used. Once capacitance is calculated, it runsthrough a couple of levels of digital filtering allowing for goodnoise immunity in different environments without sacrificingresponse time or power consumption.

Some of the major advantages comparing with the AD7147chip include an increased electrode count, a hardware con-figurable I2C address, an expanded filtering system, and com-pletely independent electrodes with auto-configuration built in.Regardless every chip can read only 12 capacitive sensors,bigger arrays can be made by connecting different MPR121chips through I2C bus. When bandwidth is not enough for

2http://www.freescale.com/webapp/sps/site/prod_summary.jsp?code=MPR121

4

Table IFEATURES OF THE MPR121 CHIP

Feature ValueNumber of electrodes 12 electrodesMeasurement range 10 pF to 2000 pF

Sensitivity range -0.01953 to -0.19531 pF / ADC countI2c bandwidth 400 Khz

Size 3x3x0.6 mm3

Figure 5. The Y1 and Cube-M families of modules

transmitting all chip readings, digital multiplexers for com-munications or other microcontrollers should be used.

IV. MECHANICS

A. Modules

Since 2004 our groups have been working on modularrobots. We have designed different modules, grouped intotwo families: Y1 and Cube-M (figure 5). The Y1 family isvery cheap and very easy to build. It is intended for fast-prototyping, testings and educational purposes. The moduleshave one DOF actuated by a low-end RC servo (futaba 3003).The electronics (not shown in the figure) includes an 8-bitPIC16F876A microcontroller. This family is composed ofthree modules: Y1, RepY1 and MY1. The Y1 consist of 6laser-cut pieces of plastic glued together. It is mainly intendedfor educational purposes as it is very cheap and easy to buildby students. RepY1 modules are the “printable” version thatcan be manufactured using a commercial 3D printer or a low-cost open Reprap machine3. The MY1 version is made ofaluminum. It has three pieces that are joined together usingscrews and nuts.

The Cube-M family[21] is an improvement over Y1. Thesemodules are made of aluminum and are designed to carry moreweight and to build bigger modular robots with 1D and 2Dtopologies. The electronics has been improved and embeddedinto the modules.

B. Robot prototypes

Some of the modular robots we have built are shownin figure 6. We have mainly concentrated on studying thelocomotion of modular robots with 1D topology. This groupcan be divided into two subgroups: pitch-pitch and pitch-yaw

3http://reprap.org/wiki/Main_Page

Figure 6. Some of the modular robot prototypes. a) Dr. Zhang andDr. Gonzalez-Gomez sitting by their modular robots b) Hypercube snake,composed of 8 pitch-yaw Y1 modules. c) Cube Revolutions: a caterpillar with8 pitch Y1 modules. d) A 2D topology star robot. e) A 5 pitch-yaw Cube-Mmodule snake. f) A four MY-1 module pitch caterpillar going through a tube.

connecting modular robots. The former is composed of thoserobots in which all the modules pitch up and down (fig. 6cand 6f). This configuration is kinematically similar to thatof the caterpillars in nature. Therefore, these robots are veryuseful for studying the climbing capabilities, even though theyonly can move forward and backward. The later comprises therobots with pitch and yaw modules. These snakes can moveon a plane (fig. 6b and 6e). When sinusoidal oscillators areused for their control, they can perform at least five differentlocomotion gaits[13]: moving in a straight line, circular path,side-winding, rotating and rolling. The relationships betweenthe oscillator’s parameters (amplitudes and phase differences)determine the type of movement that will be executed.

Modular robots with a 2D topology have also been devel-oped (fig6d) but their use for search and rescue operations isleft for future work.

C. Touch strip prototype

The first prototype of touch strip is being made and de-signed, but a preliminary proof of concept has been built. It itis shown in figure 7a. It consist of one layer with 12 electrodesin a flexible PCB, which has been taken from a commercialflexible keyboard. The strip containing the 12 electrodes hasbeen cut and three wires have been soldered for performingpreliminary tests. The total strip length is 208mm.

In figure 7b the strip is placed around one MY1 module,which has a square section. As the strip is flexible, it can befitted to any shape. There are three electrodes per side, butonly the three at the bottom are currently being tested. Theproof of concept is being done with the pitch-pitch minimumconfiguration robot consisting of two MY1 modules. This isthe minimum modular robot with locomotion capabilities[13].

5

Figure 7. Preliminary touch strip prototype. Only one layer is being tested.a) The touch strip with 12 electrodes along with the pitch-pitch minimumconfiguration robot. Only three electrodes are connected. b) The touch stripis placed around one module to form a square “touch ring”

V. EXPERIMENTS

A. Touch strip experiments

First, we have tested and measured standard capacitivesensors provided by the Freescale Sensor Toolbox MPR121Evaluation Kit shown in figure 8a. Then we have replacedthem by our touch strip prototype to test the viability of ouridea. Even if these preliminary electrodes are not meant forbeing used as capacitive sensors, because the come from aflexible commercial keyboard, they are working good enoughand the MPR121 chip is able to perform measurements. Thepressure exerted with the fingers is detected, as shown in fig.8b. As can be seen, the idea is working. These results arevery promising and let us continue to the next step: buildinganother touch strip prototype that includes the three layers andthe 12 electrodes.

B. Modular Grasping simulation

Some preliminary grasping experiments have been per-formed in simulation. The OpenRave simulator[22] has beenused along with our OpenMR4 plugin for modular robots. Thetesting program obtain the contact points between the snakeand the cylinder, as well as the applied forces. In figure 9a a

4OpenRave Modular Robots plugin: http://bit.ly/9a3fXk

Figure 8. Testing the touch strip with the Freescale Sensor Toolbox MPR121Evaluation Kit. a) The kit is connected to the standard electrodes suppliedby the manufacturer. b) Screenshot of the measures taken from the threeelectrodes of the touch strip, using the software provided by Freescale

Figure 9. Simulation of a 30 module snake robot grasping a cylinder.a) Simulation in OpenRave. b) The contact point cloud obtained from thesimulation. c) The contact point cloud view from another angle, so that thecylinder contour can be seen

6

30 module snake robot grasping the cylinder can be seen. Thecontact point cloud is shown in fig. 9b. When the orientation ofthese points is aligned with the observer, the cylinder contourcan be glimpsed.

VI. CONCLUSIONS

We have presented a novel idea for implementing the senseof touch in snake modular robots. It is based on touch stripsplaced around the snake section, forming touch rings withembedded capacitive sensors. Even if this work is in a verypreliminary stage of development, the experiments carried outconfirm the viability of this design. In addition, experimentson modular grasping have been simulated for obtaining thecontact point and forces applied when a snake robot grasp anobject.

Currently we are working in combining the locomotion,climbing and grasping capabilities with the sense of touch forthe realization of a modular snake robot for urban search andrescue operations.

VII. ACKNOWLEDGMENTS

This research is partially supported by the HANDLEproject, which has received funding from the European Com-munity’s Seventh Framework Programme (FP7/2007-2013)under grant agreement ICT 231640.

REFERENCES

[1] M. Yim, D. Duff, and K. Roufas, “Modular reconfigurable robots, anapproach to urban search and rescue,” in 1st Intl. Workshop on Human-friendly Welfare Robotics Systems, pp. 69–76, 2000.

[2] P. G. Miller, “Snake Robots for Search and Rescue,” in Neurotechnologyfor Biomimetic Robots, pp. 271–284, MIT Press, 2002.

[3] A. Wolf, H. Brown, R. Casciola, A. Costa, M. Schwerin, E. Shamas,and H. Choset, “A mobile hyper redundant mechanism for search andrescue tasks,” in Proc. of the 2003 IEEE/RSJ Int. Conf. on IntelligentRobots and Systems, vol. 3, pp. 2889–2895, 2003.

[4] S. Hirose and E. F. Fukushima, “Snakes and Strings: New RoboticComponents for Rescue Operations,” The International Journal ofRobotics Research, vol. 23, no. 4-5, pp. 341–349, 2004.

[5] Michele Guarnieri, I. Takao, E. F. Fukushima, and Shigeo Hirose, “HE-LIOS VIII: Toward Practical Robots for Search and Rescue Operations,”Journal of Robotics and Mechatronics, vol. 20, no. 5, pp. 675–694, 2008.

[6] S. Hirose and H. Yamada, “Snake-Like Robots,” IEEE Robotics &Automation Magazine, vol. 16, no. 1, pp. 88 – 98, 2009.

[7] H. Zhang, W. Wang, and Z. Deng, “A Novel Reconfigurable Robot forUrban Search and Rescue,” International Journal of Advanced RoboticSystems, vol. 3, no. 4, pp. 359–366, 2006.

[8] B. Li, S. Ma, J. Liu, M. Wang, T. Liu, and Y. Wang, “AMOEBA-I: AShape-Shifting Modular Robot for Urban Search and Rescue,” AdvancedRobotics, vol. 23, no. 9, pp. 1057–1083, 2009.

[9] J. Borenstein, M. Hansen, and A. Borrell, “The OmniTread OT-4serpentine robot-design and performance,” Journal of Field Robotics,vol. 24, pp. 601–621, July 2007.

[10] L. Rimassa, M. Zoppi, and R. Molfino, “A modular serpentine rescuerobot with climbing ability,” Industrial Robot: An International Journal,vol. 36, no. 4, pp. 370–376, 2009.

[11] Y. Changlong, M. Shugen, L. Bin, L. Hongjun, and W. Hequan,“Development of a 3D Snake-like Robot: Perambulator-II,” in Proc. of2007 IEEE Int. Conf. on Mechatronics and Automation, (Harbin, China),pp. 117–122, August 2007.

[12] J. Gonzalez-Gomez, H. Zhang, E. Boemo, and J. Zhang, “Locomotioncapabilities of a modular robot with eight pitch-yaw-connecting mod-ules,” in Proc. of the Int. Conf. on climbing and Walking machines,pp. 150–157, 2006.

[13] J. Gonzalez-Gomez, Modular Robotics and Locomotion: Application toLimbless Robot. Phd, Universidad Autonoma de Madrid, 2008.

[14] H. Zhang, W. Wang, J. Gonzalez-Gomez, and J. Zhang, “Design andRealization of a Novel Modular Climbing Caterpillar Using Low-Frequency Vibrating Passive Suckers,” Advanced Robotics, vol. 23,no. 7, pp. 889–906, 2009.

[15] G. Salvietti, H. X. Zhang, J. Gonzalez-Gomez, D. Prattichizzo, andJ. W. Zhang, “Task priority grasping and locomotion control of modularrobot,” in Proc. of the 2009 IEEE Int. Conf. on Robotics and Biomimet-ics, pp. 1069–1074, December 2009.

[16] E. R. Kandel, J. H. Schwartz, and T. M. Jessell, Principles of NeuralScience. McGraw-Hill Medical, 4th ed., 2000.

[17] G. Westling and R. S. Johansson, “Factors influencing the force controlduring precision grip.,” Experimental brain research, vol. 53, pp. 277–84, January 1984.

[18] P. Puangmali, K. Althoefer, L. D. Seneviratne, D. Murphy, and P. Das-gupta, “State-of-the-Art in Force and Tactile Sensing for MinimallyInvasive Surgery,” IEEE Sensors Journal, vol. 8, pp. 371–381, abril2008.

[19] P. A. Schmidt, E. Mael, and R. P. Waurtz, “A sensor for dynamic tactileinformation with applications in human-robot interaction and objectexploration,” Robotics and Autonomous Systems, vol. 54, no. August2006, pp. 1005–1014, 2006.

[20] M. Maggiali, G. Cannata, P. Maiolino, G. Metta, M. Randazzo, andG. Sandini, “Embedded Distributed Capacitive Tactile Sensor,” in Proc.of the 11th Mechatronics Forum Biennial Int. Conf., no. 039, pp. 1–5,2008.

[21] H. Zhang, J. Gonzalez-Gomez, S. Chen, and J. Zhang, “EmbeddedIntelligent Capability of a Modular Robotic System,” in Proc. of theIEEE Int. Conf. on Robotics and biomimetics, (Bangkok, Thailand),pp. 2061–2066, IEEE Computer Society, 2009.

[22] R. Diankov and J. Kuffner, “OpenRAVE: A Planning Architecturefor Autonomous Robotics,” Tech. Rep. CMU-RI-TR-08-34, RoboticsInstitute, 2008.